deploy guard: fix actions

.gitea/workflows/deploy.yml

@@ -63,6 +63,13 @@ jobs:
       - name: Build muffin
         run: nix build .#nixosConfigurations.muffin.config.system.build.toplevel -L
 
+      - name: Deploy guard preflight
+        run: |
+          ssh -i /run/agenix/ci-deploy-key \
+            -o StrictHostKeyChecking=yes \
+            -o UserKnownHostsFile=/etc/ci-known-hosts \
+            root@server-public deploy-guard-check
+
       - name: Deploy via deploy-rs
         run: |
           eval $(ssh-agent -s)

AGENTS.md

@@ -158,10 +158,12 @@ Hard requirements that are asserted at eval time:
 
 ## Deploy guard (muffin)
 
-`modules/server-deploy-guard.nix` blocks `./deploy.sh muffin` / deploy-rs activation when a service it covers is in active use. Two paths enforce it:
+`modules/server-deploy-guard.nix` aggregates per-service "is anyone using this right now?" checks into a single `deploy-guard-check` binary on muffin. Enforcement is **preflight-only** — the guard runs over SSH *before* deploy-rs is invoked; activation itself is never gated. This matters because deploy-rs sets the new profile pointer before running the activation script, so a failed activation triggers auto-rollback which re-runs `switch-to-configuration` on the previous generation — that re-activation rotates agenix secrets, reinstalls lanzaboote, and reloads systemd units. The only safe place to stop a deploy is before deploy-rs starts.
 
-- **Preflight**: `./deploy.sh muffin` SSHes to `server-public` and runs `deploy-guard-check` before the build. Connectivity failure is soft (activation still enforces). `./deploy.sh muffin --force` or `DEPLOY_GUARD_FORCE=1 ./deploy.sh muffin` touches `/run/deploy-guard-bypass` remotely (single-shot) and skips the preflight.
+Two drivers invoke the preflight:
-- **Activation**: the custom `activate.custom` wrapper in `flake.nix` runs `$PROFILE/sw/bin/deploy-guard-check` before `switch-to-configuration switch`. A non-zero exit triggers deploy-rs auto-rollback. Same bypass: `DEPLOY_GUARD_BYPASS=1` env or pre-touched `/run/deploy-guard-bypass`.
+
+- **`./deploy.sh muffin`** SSHes to `server-public` and runs `deploy-guard-check`. SSH connection failure is a hard abort (rc=255) because there is no second gate. `./deploy.sh muffin --force` (or `DEPLOY_GUARD_FORCE=1 ./deploy.sh muffin`) skips the preflight entirely.
+- **CI (`.gitea/workflows/deploy.yml`)** has a `Deploy guard preflight` step between `Build muffin` and `Deploy via deploy-rs`. A non-zero exit fails the job before any closure copy or activation.
 
 ### Adding a new check
 

deploy.sh

@@ -6,8 +6,8 @@
 #   ./deploy.sh switch            # apply immediately on current host
 #   ./deploy.sh test              # apply without adding boot entry
 #   ./deploy.sh build             # build only, no activation
-#   ./deploy.sh muffin            # build + deploy to muffin via deploy-rs
+#   ./deploy.sh muffin            # preflight deploy guard + deploy-rs to muffin
-#   ./deploy.sh muffin --force    # bypass the deploy guard (active-user check)
+#   ./deploy.sh muffin --force    # skip the preflight deploy guard
 #
 # muffin cannot be rebuilt locally from another host — this script only issues
 # the remote deploy via deploy-rs when explicitly named.
@@ -28,27 +28,26 @@ case "$arg" in
     if [ "${1:-}" = "--force" ]; then force=1; shift; fi
 
     if [ "$force" = "1" ]; then
-      echo "deploy-guard: bypass requested; setting remote marker"
+      echo "deploy-guard: preflight skipped (--force)"
-      ssh -o BatchMode=yes -o ConnectTimeout=3 root@server-public \
-        'touch /run/deploy-guard-bypass' \
-        || echo "deploy-guard: warning: could not write remote bypass marker" >&2
     else
-      # Single SSH probe — if exit 255 it's a connectivity failure (skip
+      # Single SSH probe. Exit 255 is a connectivity failure; treat as a hard
-      # preflight; the activation-time guard still enforces). Any other
+      # abort — without the preflight there is no other gate that prevents
-      # non-zero is the guard blocking the deploy.
+      # deploy-rs from partially activating while users are online.
       output=$(ssh -o BatchMode=yes -o ConnectTimeout=5 \
         root@server-public deploy-guard-check 2>&1) && rc=0 || rc=$?
 
-      if [ "$rc" -eq 255 ]; then
+      if [ "$rc" -eq 0 ]; then
-        echo "deploy-guard: muffin unreachable for preflight;" \
+        [ -n "$output" ] && printf '%s\n' "$output"
-             "activation will still enforce" >&2
+      elif [ "$rc" -eq 255 ]; then
-      elif [ "$rc" -ne 0 ]; then
+        echo "deploy-guard: preflight SSH failed (rc=255)." >&2
+        printf '%s\n' "$output" >&2
+        echo "Re-run with --force once you've confirmed the host is idle." >&2
+        exit 1
+      else
         printf '%s\n' "$output"
         echo >&2
         echo "Blocked by deploy guard. Bypass: ./deploy.sh muffin --force" >&2
         exit 1
-      elif [ -n "$output" ]; then
-        printf '%s\n' "$output"
       fi
     fi
 

flake.nix

@@ -372,37 +372,17 @@
         profiles.system = {
           sshUser = "root";
           user = "root";
-          # Wrap deploy-rs.activate.nixos so the guard runs before
+          # Deploy guard enforcement lives in the preflight driver (deploy.sh
-          # switch-to-configuration. If the guard exits non-zero, deploy-rs's
+          # and .gitea/workflows/deploy.yml) — not in activation. Activation-
-          # auto-rollback restores the previous profile. Bypass via
+          # time enforcement is unsafe: deploy-rs sets the new profile pointer
-          # DEPLOY_GUARD_BYPASS=1 or by pre-touching /run/deploy-guard-bypass.
+          # before running deploy-rs-activate, so a non-zero activation exit
-          path =
+          # triggers auto-rollback which re-runs switch-to-configuration on the
-            let
+          # previous generation. That re-activation rotates agenix secrets,
-              base = self.nixosConfigurations.muffin;
+          # reinstalls lanzaboote, and reloads systemd units — side effects we
-              activate = deploy-rs.lib.${system}.activate;
+          # want to avoid when the deploy is supposed to be a no-op blocked by
-              bootloaderDefaultCleanup = nixpkgs-stable.lib.optionalString base.config.boot.loader.systemd-boot.enable "sed -i '/^default /d' ${base.config.boot.loader.efi.efiSysMountPoint}/loader/loader.conf";
+          # the guard. Blocking before the deploy-rs invocation is the only
-            in
+          # clean way to leave the running system untouched.
-            (
+          path = deploy-rs.lib.${system}.activate.nixos self.nixosConfigurations.muffin;
-              activate.custom
-              // {
-                dryActivate = "$PROFILE/bin/switch-to-configuration dry-activate";
-                boot = "$PROFILE/bin/switch-to-configuration boot";
-              }
-            )
-              base.config.system.build.toplevel
-              ''
-                # work around https://github.com/NixOS/nixpkgs/issues/73404
-                cd /tmp
-
-                # Halt deploys while users are actively using services.
-                # See modules/server-deploy-guard.nix.
-                "$PROFILE/sw/bin/deploy-guard-check"
-
-                $PROFILE/bin/switch-to-configuration switch
-
-                # https://github.com/serokell/deploy-rs/issues/31
-                ${bootloaderDefaultCleanup}
-              '';
         };
       };